Аннотация:We consider a signal segmentation problem within the hidden Markov model (HMM) approach and try to take
into account label frequency constraints. Following the dual decomposition approach we maximize an energy lower bound via subgradient ascent method, where subgradient is found on each iteration by solving two subproblems. The first subproblem can be effectively solved by Viterbi algorithm and the other one can be reduced to an easy-to-solve transportation problem. We show the efficiency of our approach on toy signals and on the task of automated segmentation of mouse behavior.